py3: flush output from FLINT error message

[jhpalmieri]

This is a second fix for the problem reported at #28334, the doctest failure in rings/polynomial/polynomial_rational_flint.pyx. This fix does not rely on patching FLINT. See ​sage-devel discussion.

[jhpalmieri]

As I said at #28334, the fix there can be kept independently of the change here, and it should be kept, since the patch to FLINT was accepted upstream. The one here is to handle the case when using non-patched versions of FLINT.

[jhpalmieri]

New commits:

​cd89b9d

trac 28649: flush C buffer

[dimpase]

good, thanks. Takes extra 0.5 sec on my laptop, but OK...

[nbruin]

It may be possible to do this cross-platform in a relatively convenient way with ctypes anyway:

import ctypes
libc = ctypes.CDLL(ctypes.util.find_library("c"))
libc.fflush(0r)

seems to work, and looking at the source, it seems "find_library" does make an effort to be cross-platform. It's probably worth considering if this will do, since it's a lot lighter weight than running cython.

[dimpase]

we can just add this flush function somewher into sagelib

[jhpalmieri]

Using Sage's own misc.compat.find_library would probably be better:

src/sage/rings/polynomial/polynomial_rational_flint.pyx

@@ -1176 +1176 @@
             ...
             RuntimeError: FLINT exception
 
+        Flush the output buffer to get rid of stray output -- see
+        :trac:`28649`::
+
+            sage: from sage.misc.compat import find_library
+            sage: import ctypes
+            sage: libc = ctypes.CDLL(find_library("c"))
+            sage: libc.fflush(0r)
+            ...
+
         Test fractional powers (:trac:`20086`)::
 
             sage: P.<R> = QQ[]

[jhpalmieri]

Both approaches work for me on OS X and on my Ubuntu virtual machine. Should they both work on Cygwin?

[nbruin]

It looks like

libc = ctypes.CDLL(None)

works too. I have no idea why. (see ​https://eli.thegreenplace.net/2015/redirecting-all-kinds-of-stdout-in-python/)

[jhpalmieri]

Does ​https://stackoverflow.com/questions/49878901/how-does-ctypes-cdll-loadlibrarynone-work help? I'm worried about this on Cygwin, though.

[jhpalmieri]

Any opinions on which approach to take? I see three options:

1. Keep the current branch. Slow, requires a functioning Cython compiler. (Is that guaranteed with binary releases?)

2. Add the cyflush function to the Sage library so that it is precompiled.

3. Take the ctypes approach from comment:17, with libc = ctypes.CDLL(None).

Any opinions? I would like to get a solution to this problem merged, because there are people out there using non-Sage versions of FLINT.

[dimpase]

I think 2 is the easiest. I don't see any benefits of using ctypes here, whatsoever.

[jhpalmieri]

Any suggestions for where that function should go? polynomial_rational_flint.pyx is one option, since it will be used there, but that's not a particularly natural place for anyone to find it later.

[chapoton]

src/sage/misc/pager.py ?

[chapoton]

or maybe misc/sage_ostools.pyx

[jhpalmieri]

I'm leaning toward either misc/misc_c.pyx or a new file cpython/cyflush.pyx.

[git]

Branch pushed to git repo; I updated commit sha1. This was a forced push. New commits:

​461e017

trac 28649: flush stray output from external library calls

[jhpalmieri]

Let's try this.

[git]

Branch pushed to git repo; I updated commit sha1. This was a forced push. New commits:

​bc01c65

trac 28649: flush stray output from external library calls

[chapoton]

the EXAMPLES:: content in cyflush is overindented.

[git]

Branch pushed to git repo; I updated commit sha1. This was a forced push. New commits:

​9957855

trac 28649: flush stray output from external library calls

[jhpalmieri]

Replying to chapoton:

the EXAMPLES:: content in cyflush is overindented.

Oops, sorry. Fixed.

[chapoton]

ok, good to go